Sonic analysis in my non-classical music library finished last night. I’m looking through some artists to see what it came up with, and I found some non-intuitive results. The first example is Herb Alpert. I have a couple albums, plus a few tracks, under artist “Herb Alpert.” I also have a greatest hits album under artist “Herb Alpert & The Tijuana Brass.” But sonic analysis found “Herb Alpert” to be sonically similar to “Spyro Gyra,” “Grover Washington, Jr.,” and “Diana Krall.”
I would think that Herb Alpert & TJB would be in that list, at least more likely than Diana Krall.
Maybe “under the hood,” there’s some valid reason to find Herb Alpert and Diana Krall to be sonically similar, but practically, intuitively, I just don’t see it.
This is reminiscent of Plex’s auto-tagging of photos, where it decided my waterfalls with moss-covered rocks were “broccoli.”
Other oddities:
George Benson sonically similar to Timothy B Schmit
Michael Jackson sonically similar to Nu Shooz
Donald Fagen not sonically similar to Steely Dan
Don Henley not sonically similar to The Eagles
Mark Knopfler not sonically similar to Dire Straits or The Notting Hillbillies
I don’t know if I’ll have much use for this feature.
You have Donald Fagen similar to Steely Dan, but not sonically similar, so our results are similar there, too.
I have a lot of artists with no sonically similar entries. My library isn’t overly large compared to what some have mentioned of theirs, so I’m not particularly surprised at that. But these seemingly inaccurate examples don’t bolster my interest in the feature (not that I was overly excited to begin with).
At least your scan finished. I think my library is scan is broken. It was at 2,090 albums remaining yesterday, but today it jumped up to 2,108 and has been pegged there all day.
(As I’ve said in past threads, if you’re truly interested in “exploring” your library of music, you can’t beat the features offered by Roon. It’s expensive though, so probably a point of resistance for many.)
In some ways that might be the point of the sonically similar artists feature. To get away from artists that we intrinsically associate with each other, and associate artists by reference to how their music sounds.
As a comparison data point, Don Henley is sonically similar to the Eagles on my analysis. But Glenn Frey is not, which is not entirely surprising because Glenn Frey’s solo albums did not sound at all like Eagles albums. Even then, Don Henley and Glenn Frey appear as sonically similar which probably reflect that most of their solo material was produced at the same time (80s) when there would have been similar production styles for those two artists.
An interesting thing I have noted is an almost complete separation of male and female vocalists, at artist, album and track level. The sonic mixes/radios I have listened to have been very nearly exclusively male or female vocalists. There has probably only been one or two crossovers.
But I just wonder whether the analysis places too much weight on vocals? There are many male and female artists with similar musical styles but they don’t match which I can only presume is the result of vocal differences. That may be unavoidable in the analysis.
The way I look at it is that sonic mixes will avoid the jarring experience you get when The Everly Brothers are followed by Metallica. So, for example, I wouldn’t be jarred by the Eagles following Linda Ronstadt. Or Joan Baez following Bob Dylan. But those artists don’t sonically match.
Yes results can be quite odd. I’ve seen in the release notes from the last update that there is some fixing about sonic analysis maybe that will fix some things ?
Here is an exemple that is the album page from a classic reggae / dub album from Junior Murvin, produced by Lee “Scratch” Perry, that’s the only album I have from this artist.
On the album page the only sonically similar albums are “We are Familly” and “Thriller”.
I’m seeing strange results as well, and inconsistencies.
For example, earlier today Seldom Scene’s “The Hometown Blues” was rated sonically similar to Armin Van Buuren’s “Beautiful Life (Kat Krazy Radio Edit)”, which is…mind bogglingly strange.
But just now, the same Seldom Scene song is no longer sonically similar to Armin Van Buuren, but a more logical Infamous Stringdusters version of “Deep Elem Blues.”
I have a very diverse and eclectic collection, so I’ve been really looking forward to some interesting matches. But some are apparently random. I’d love to see data on why it makes certain matches.
I’m facing the same questions. For instance, I have some Brazilian regional music albums that were classified similar to Elton John. There’s nothing even close to their music styles. While other albums that are also Brazilian regional music were not related to each other, even though they had the same genre tag. Of course, I want to discover not so obvious relations, but first things first: those that clearly have the same style are being classified w/o any similarity, when I expected they were classified as similar.
Btw, how can I have plex server re-analyze the library for sonic classification?
